1. Field of the Invention
The present invention relates generally to solid state image sensors, and more specifically to a novel three-dimensional image sensor structure.
2. Description of the Related Art
Visible imaging systems produced with CMOS image sensors significantly reduce camera cost and power while improving image resolution and reducing noise. The CMOS image sensors are typically imaging System-on-Chip (iSoC) products that combine image detection and signal processing with a host of supporting intellectual property (IP) blocks including timing controller, clock drivers, reference voltages, A/D conversion, image processing stages, and other ancillary circuits. Consequently, the resulting video cameras can be assembled using a single CMOS integrated circuit supported by only a lens, shutter and a battery. The result is smaller and smaller cameras with longer and longer battery life at ever lower cost.
The improvements delivered by CMOS iSoC sensors, including especially the operational flexibility enabled by their embedded iSoC functionality, have also translated to the emergence of dual-use cameras that produce both high-resolution still images and high definition video. This convergence of still capture and video acquisition has obsolesced both dedicated still cameras and conventional camcorders produced with prior sensor technologies, such as CCDs. It has also exposed the need for even better dual-use image sensors to optimally perform both types of imaging.
While CMOS iSoC dual-use sensors produce both stills and video exhibiting acceptable quality for many applications, their image quality is well below the limit set by device physics. Furthermore, the capture quality is usually somewhat degraded under diverse lighting conditions and severely compromised in extreme conditions.
One example of a challenging scenario is taking a picture of a tree that is directly backlit by the sun; the combination of shade, numerous specular highlights on illuminated leaves, and direct sunlight nearly always results in sub-optimal image quality. A video capture of the backlit tree is even more challenging, especially if wind is combined with variable cloud cover; this mélange of shade, mid-tones and specular highlights is exceptionally difficult to capture with optimum fidelity. Further adding to the challenge, such circumstances frequently and dynamically emerge within seconds of relatively uniform and benign lighting conditions. The challenge of capturing “perfect” images is further complicated by the fact that photographers gravitate toward the most challenging lighting conditions to maximize photo aesthetics (by leveraging the so-called “magic hour” at sunrise and sunset).
The variable and volatile scene dynamics impact final image quality not only during the exposure time, but also at all other times due to the fact that many elements of the image sensor are photosensitive via direct or indirect means. This parasitic signal capture generates imaging artifacts that degrade image quality. Sensors with internal electronic shutters cannot prevent much of the stray signal from contaminating the image capture. Inclusion of a mechanical shutter usually helps prevent most of the parasitic signal generation from occurring. However, including a mechanical shutter adds cost, complexity, and decreases camera reliability; there is consequently a compelling need to eliminate its inclusion.
Nevertheless, the best means developed to date for truly blocking light is a mechanical shutter; the resulting shutter rejection ratio (SRR) can be close to infinity, i.e., no light impinging on the camera is detected anywhere in the sensor when the shutter is closed. The detection does not have to be on the actual photodetector, but can instead be picked up elsewhere in various circuits to affect performance. The sensor's shutter rejection ratio is also often referred to as the extinction coefficient describing its ability to electronically block light during the period when light capture is disabled, i.e., the epoch when a mechanical shutter would be placed in front of the sensor so that zero unwanted signal is collected.
Monolithic sensors with electronic shutters are not as blind to ambient light as when mechanical shutters are used. Nevertheless, in order to further reduce costs, camera manufactures wish to eliminate mechanical shutter mechanism for still cameras by having sensor manufactures supply devices delivering extremely high SRR. CMOS iSoCs hence need to have an SRR that is well in excess of 100 dB, which is well beyond the extinction limits of modern CMOS and CCD image sensors.
One approach for eliminating the mechanical shutter is to produce image sensors having an electronic rolling shutter. The image is formed in these sensors on a line-by-line basis such that there is always a delay of one frame time from the starting/ending of the first line's exposure to the starting/ending of the last line's exposure. The result is that each line effectively captures a different epoch. Whether for still capture or video, highly objectionable artifacts can ensue for capture rates below about 60 Hz, depending on the rate of movement in the scene. On the other hand, the overall performance of rolling shutter sensors is generally superior to sensors with global shutter capability, wherein the entire sensor captures an identical exposure epoch, because the pixels are simpler to design and build; the signal-to-noise ratio of rolling shutter sensors is vastly superior to those with global shutter.
The mechanical shutter can otherwise be replaced by integrating an electronic global shutter in the image sensor. In this type of sensor each pixel integrates its captured signal during a single, identical exposure period. Despite having a much more complex pixel design, the sensor must perform without compromise, such that the measured performance is very high and not limited by device design or process technology. Up to now, CMOS global shutter sensors have exhibited lower fill factors and higher noise levels than competing CMOS rolling shutter sensors. Using system-on-chip integration to produce these “snapshot” sensors has not yet closed the gap.
The advantages offered by system-on-chip integration in CMOS visible imagers for emerging camera products have thus spurred considerable effort to further improve active-pixel sensor (APS) devices by developing high performance global shutter functionality. Unfortunately, in addition to higher noise, inferior fill factor, and vulnerability to parasitic signal pickup, the increasingly sophisticated iSoCs are also more vulnerable to noise pickup. The undesirable pickup is especially likely within the most desired sensors: the mode-changing sensors capable of high quality still and video capture. One objectionable result is increased noise, both coherent clock feedthrough and fixed pattern noise, because dual-mode usage dynamically changes the sensors self-EMI and clock feedthrough, thereby variably impacting image quality.
Current image sensor designs with snapshot image capture capability hence still require a mechanical shutter to most effectively perform correlated double sampling (CDS), wherein a first dark frame is subtracted from a second exposed frame in order to eliminate the sensor's reset (or kTC) noise while also reducing fixed pattern noise. In the absence of the mechanical shutter, the various parasitic signals increase the post-CDS noise to well above the fundamental limit normally set by the time interval between the frame subtractions.
Modern image sensor designs with rolling shutter image capture capability work more effectively without a mechanical shutter since the rolling shutter electronic circuits can be used to minimize the dead time during which some of the sensor's circuitry is vulnerable to light contamination by either direct or indirect means. Dead time minimization via prudent rolling shutter timing hence limits the integration of unwanted signal throughout the sensor, including the many other photosensitive locations outside each pixel's photodetector.
A final major drawback of the embedded on-chip CMOS circuit sophistication for producing high performance image sensors is that these devices are invariably produced using CMOS technology that was developed by starting with “standard” CMOS processes that were modified to subsequently add imaging aspects. These ex post facto modifications re-engineered the CMOS technology for imaging even though the underlying technology was originally optimized for producing largely digital systems-on-a-chip.
A first outcome is that these CMOS “image sensor” (CIS) processes have many mask layers, thereby increasing the costs associated with manufacturing.
A second outcome is that the resulting CMOS imaging processes offer digital logic at technology nodes well behind the state of the art such that the benefits of Moore's law have not been fully exploited in CMOS image sensors.
A final outcome, which the empirical results over the last decade incontrovertibly show, is that it has not been possible to truly optimize photodiode quality in these monolithic CIS processes; while the mean dark current is roughly comparable to that routinely achieved with commercial CCD production, the dark current is higher when compared to scientific CCDS and, most unbearable, the number of defective pixels is several orders of magnitude larger. The integrated processes integration is therefore still better suited for digital logic rather than the more delicate photodetector. This deficiency is not surprising since only recently has there been sufficient production demand for high-quality sensors to justify specifically developing an optimized CIS process at the world's semiconductor foundries.
Whereas developing an optimized CMOS image sensor process would require very expensive semiconductor process development targeted specifically for image sensors that would have vastly different requirements relative to the mainstream consumer-driven technologies still driving much larger production volumes, the present invention delivers a more tractable solution.